Abstract: Based on the features of gestures and regression algorithms, by improving the defects of the traditional 3D gesture cascading regression algorithm, we proposed a new operator of handstructure features based on the gestural variations. This operator efficiently reduced the dimension of handfeature operators. We adapted to the structural improvement of traditional cascading regression in terms of the changes of gestures, so that it significantly improved the recognition efficiency aiming at the handy structures. The experiments were carried out on the public databases and selfbuilt databases, and the experimental results indicates that the algorithm can successfully improve the execution efficiency while maintaining the accuracy of gesture recognition.

Key words: cascaded regression, gesture structure, gesture